b5df59660f
don't set pcap timeout, to prevent cpu churn pcap timeouts are a mystery. Setting the timeout to -1 used to reduce latency, but that was before we started setting immediate mode. And it turns out that with that timeout, pcap chews through a lot of CPU polling its file descriptor, checking timers, etc.
Matthias Radestock2014-09-23 11:06:41 +01:00
eecaa567fc
eliminate periodic 'tick' in connection_maker checkStateAndAttemptConnections already performs a full scan of all potential targets, from which we can easily determine the nearest future time (or, rather, its duration from now) on which an attempt is due.
Matthias Radestock2014-09-22 23:15:54 +01:00
c48ae19c55
Tidy up messages; ensure errors to to stderr
Bryan Boreham2014-09-18 11:47:47 +01:00
a75a0d8709
If and only if network setup because the container died, say so. This involves capturing stderr output from all the 'ip' commands so we can hide that output if we see the process died, or show it to the user if the process didn't die, in which case we don't know why we failed.
Bryan Boreham2014-09-18 11:24:08 +01:00
bfd5007339
Don't test if symlink exists before 'rm -f' When the symlink's target doesn't exist, '[ -f' is false, so the symlink doesn't get removed. Since we do always want it to be removed, just call 'rm -f' anyway, which won't complain if the file doesn't exist.
Bryan Boreham2014-09-18 11:03:06 +01:00
a94ae91921
Merge pull request #63 from bboreham/29_unknown_peer
rade2014-09-17 15:30:53 +01:00
914eceaa71
Merge pull request #62 from christianberg/patch-1
rade2014-09-17 14:04:07 +01:00
8bac04d080
Remove message about unknown peer from the log
Bryan Boreham2014-09-17 13:19:52 +01:00
c8876e144f
Revert "Change iteration over connection decoding to abort after the first error, and let the UnknownPeersError know which peer was unknown"
Bryan Boreham2014-09-17 13:18:33 +01:00
4e7aa85de0
Revert "Make message for UnknownPeersError more meaningful"
Bryan Boreham2014-09-17 13:18:07 +01:00
a25ccbb261
Merge branch 'master' into 29_unknown_peer
Bryan Boreham2014-09-17 12:59:30 +01:00
e287110099
improve error messages and make sure they all go to stderr ...since otherwise they get hidden when the output is redirected, e.g. in the typical
Matthias Radestock2014-09-17 06:46:02 +01:00
752cfdcb67
Make message for UnknownPeersError more meaningful
Bryan Boreham2014-09-16 11:11:54 +01:00
978d374820
Change iteration over connection decoding to abort after the first error, and let the UnknownPeersError know which peer was unknown
Bryan Boreham2014-09-16 09:32:53 +01:00
5490826666
attach command that accepts and existing container id and runs configure_container_networking using it
Kai Davenport2014-09-15 10:25:59 -07:00
2e66694355
Merge pull request #58 from bboreham/50_show_ip
rade2014-09-15 18:04:01 +01:00
464a1c8246
Show the IP address on the end of each connection
Bryan Boreham2014-09-15 17:50:04 +01:00
af94153d96
Merge pull request #44 from bboreham/23_retry_peers
rade2014-09-15 17:42:32 +01:00
3e0c0d6e55
Merge pull request #57 from bboreham/56_display_version
rade2014-09-15 16:53:24 +01:00
f57058f627
Use git commit hash instead of requiring user to enter a version tag on 'make publish'
Bryan Boreham2014-09-15 15:33:09 +01:00
bd45c7212f
always notify ConnectionMaker of terminated connections otherwise CM might have them stuck in the 'attempting' state
Matthias Radestock2014-09-15 15:28:11 +01:00
500bda596c
Don't do the print args thing in this branch
Bryan Boreham2014-09-15 14:42:39 +01:00
18f65dc344
Add 'publish' target to tag and push release
Bryan Boreham2014-09-15 14:40:57 +01:00
244188776f
Add tag to build command, so you can do "make tag=x.xx" to add a version number to the image
Bryan Boreham2014-09-15 14:26:13 +01:00
39003233ea
Remove version/build-date info from weaver binary
Bryan Boreham2014-09-15 14:05:15 +01:00
5b18864aec
refactoring - rename addToTargets to addTarget - immediately add to cm.targets; the phasing was unnecessary - don't normalise addresses in addTarget - they are already normalised
Matthias Radestock2014-09-15 13:51:57 +01:00
8297fef53d
remove WEAVE iptable chain on 'weave stop' No point keeping it around.
Matthias Radestock2014-09-15 12:03:29 +01:00
2f1359ae9f
remove masquerading rules on 'weave stop/launch' Leaving iptable masquerading rules behind is a bad idea when weave is re-launched with different ip subnet settings. So now we remove the rules on 'weave stop' and also on 'weave launch' (in case weave had been stopped in ways other than 'weave stop').
Matthias Radestock2014-09-15 11:34:57 +01:00
f10c460f48
leave networking alone on 'weave launch' unless we start container
Matthias Radestock2014-09-14 19:31:27 +01:00
4c93991651
ignore conntrack result in 'weave stop' so we exit with status 0, since not having any existing flows is perfectly normal
Matthias Radestock2014-09-14 19:08:00 +01:00
948312f03d
refactor: extract container network configuration and use more function parameters in preference to global vars
Matthias Radestock2014-09-13 21:45:42 +01:00
c845ba7872
remove detritus we haven't depended on pipework for a while
Matthias Radestock2014-09-13 20:48:49 +01:00
6eb7e1aa43
disable offloading on weave bridge since w/o it the kernel just ignores the PMTU when the host talks to weave via 'weave expose'
Matthias Radestock2014-09-13 19:58:54 +01:00
Matthias Radestock